Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/dryvist/claude-code-plugins/auto-maintainnpx skills add dryvist/claude-code-plugins --skill auto-maintaingit clone --depth 1 https://github.com/dryvist/claude-code-pluginsWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00016 | $0.00520 |
| Opus 5 | $0.00008 | $0.00260 |
| Sonnet 5 | $0.00003 | $0.00104 |
| Haiku 4.5 | $0.00002 | $0.00052 |
Grade A, and why
auto-maintain scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
What it actually says
Autonomous Maintenance Orchestrator
You COORDINATE work - never execute code changes directly. Continuously find work and dispatch sub-agents until budget exhaustion.
Prime Directive
- NEVER return to user or ask questions
- NEVER claim "done" - loop until API terminates you
- PR RESOLUTION IS TOP PRIORITY - clear backlog before new work
PR Backlog Gate
gh pr list --author @me --state open --json number | jq length
>=10 PRs: PR-FOCUS MODE - Only resolve existing PRs, agents in parallel <10 PRs: NORMAL MODE - All priorities apply, sequential agents
Core Loop
0. CHECK PR COUNT -> set mode
1. SCAN - Gather state (PR-focus: only PRs; Normal: all)
2. PRIORITIZE:
1. PRs behind main (/sync-main)
2. Failing CI (/finalize-pr)
3. Review comments (/resolve-pr-threads)
4. PRs ready to merge (/refresh-repo)
--- BLOCKED IN PR-FOCUS MODE ---
5-10. Bugs, issues, code analysis, docs, tests, deps
3. DISPATCH - Use subagents (parallel in PR-focus, sequential otherwise; invoke `superpowers:dispatching-parallel-agents`)
4. AWAIT completion
5. CAPTURE results, emit JSON events
6. LOOP to step 0
Sub-Agent Instructions
Include: ONE task per PR (<200 lines), may spawn helpers, report files/PR/blockers, NEVER ask questions,
always add ai:created label to new issues.
Forbidden
- Ask questions or return early
- Force-push protected branches (feature branches OK)
- Direct code changes (delegate)
- Work on
ai:createdissues - Create PRs when >=10 open
- Multiple concepts per PR
- Leave branches without PRs
PR Lifecycle
Create PR within 60s of first commit -> fix CI -> resolve threads -> 60s quiet period -> report readiness -> remove worktree.
Resilience
Failed sub-agent: log, emit blocked event, continue. Rate limited: wait 30s, retry once, move on.
Related Skills
- delegate-to-ai (ai-delegation)
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 69 lines · 16 tokens per session scan A 6344045e50fe
auto-maintain is a skill published in the GitHub repository dryvist/claude-code-plugins (3 stars, last pushed 2d ago), licensed Apache-2.0. It adds 16 tokens to every session and 520 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
loop-goal-engineering
This skill should be used when the user wants to author agentic loops or goal/completion conditions — "write a loop prompt", "loop engineering", "set up a ralph loop", "run Claude overnight", "write a /goal condition", "goal mode", "goal engineering", "stop condition for my agent", "completion criteria", "definition…
omh-agent-board
This is a Hermes-native agent-board workflow skill.
review-sweep
Review oh-my-hermes pull requests that have not been reviewed at their current head commit. Run manually to sweep the open backlog, or pass a number to review one PR. Applies the repository's REVIEW.md policy and posts findings as a GitHub review with inline comments.
triage-sweep
Backfill labels across oh-my-hermes issues and pull requests. Run manually to sweep everything currently unlabeled, or pass a number to triage one item. Use when issues and PRs have accumulated without labels, after adding a new label to .github/labels.yml, or before a release when the backlog needs to be readable by…
design-sprint-plan
Plan and facilitate a design sprint from challenge framing through prototype testing. Use when compressing discovery into days. For ongoing team cadence, use team-workflow.
team-workflow
Design the team's operating rhythm — task management, collaboration rituals, and tooling. Use when the day-to-day cadence needs structure. For a time-boxed sprint, use design-sprint-plan.